Explainable machine learning model for predicting short-term outcomes in sepsis- induced coagulopathy.

Area of Science:

• Critical Care Medicine
• Medical Informatics
• Computational Biology

Background:

• Sepsis-induced coagulopathy (SIC) is a frequent complication of sepsis.
• SIC is associated with increased mortality risk in sepsis patients.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting 28-day mortality in patients with SIC.
• To identify key predictors of mortality in SIC patients.

Main Methods:

• Data from the MIMIC-IV database was used for model training and external validation.
• Least Absolute Shrinkage and Selection Operator (LASSO) regression and logistic regression identified predictive factors.
• XGBoost classification model was developed and evaluated using ROC curves, calibration curves, and DCA.
• SHAP values were used for model interpretability.

Main Results:

• The XGBoost model achieved an AUC of 0.840 on the test set, with 80.7% accuracy.
• External validation showed excellent performance with an AUC of 0.864.
• The model demonstrated strong predictive power for 28-day mortality in SIC patients.

Conclusions:

• An interpretable XGBoost model was successfully developed to predict 28-day mortality in SIC patients.
• The model can support clinical decision-making and personalized treatment strategies.
• This tool offers a valuable basis for assessing mortality risk in SIC.